IEEE survey covers neuromorphic event cameras in robotics with human-in-the-loop scenarios

Applications of Neuromorphic/Event Camera in Robotics With Human in Loop: A Systematic Review, Datasets, and Challenges (IEEE Xplore, 2025) surveys event camera applications in robotic systems where a human operator remains in the decision loop — receiving sensor information, making high-level decisions, but delegating low-level perception and actuation to the machine.

The human-in-the-loop framing

The human-in-the-loop (HITL) framing is important for defense applications because it corresponds to the doctrinal constraint on autonomous weapons: humans maintain decision authority over lethal effects. In this model, the sensor system handles detection, classification, and tracking autonomously, but the engagement decision requires human authorization.

This architecture places different requirements on the sensing system than either fully autonomous or fully human-operated systems. The sensor must produce reliable, legible output that enables confident human decision-making quickly, because the decision cycle is the bottleneck. False alarms that require human evaluation waste the decision cycle; missed detections put the human in an engagement scenario without information.

What event cameras contribute to HITL robotics

The review identifies three areas where event cameras improve HITL systems:

Alert latency. In surveillance applications, the time from when a threat appears in the sensor’s field of view to when the human operator receives an alert is a critical metric. Frame-based systems add at least one frame period (33 ms at 30 fps) before any detection can begin. Event-based systems produce measurements within microseconds of the target’s first appearance. This matters when the human decision cycle is measured in seconds and the alert must trigger early enough for the human to act.

Operator workload. HITL systems fail when they overwhelm the operator with more alerts than can be evaluated. Event-based detection, by being motion-selective, generates alerts primarily when scene changes occur, which reduces the rate of false alerts compared to frame-based detection that must analyze every frame whether or not anything is happening.

Degraded conditions. Human operators become unreliable in degraded-sensor conditions — low light, motion blur, high clutter — before the machine does. Event cameras extend the condition envelope where the sensor can provide usable information to the human.

Dataset gaps

The review identifies that most available event-camera datasets simulate the sensor’s view, not the human operator’s view. Datasets that capture both the sensor output and the human operator’s decisions and decision times are scarce. Without such data, it is difficult to optimize the sensor-human interface.

This is a general gap in HITL robotics research that has not been addressed specifically for event cameras, and it matters for practical deployment: a sensor system’s effectiveness in a HITL scenario depends as much on the quality of the human-machine interface as on the sensor performance metrics.

Applicability to C-UAS

Semi-autonomous C-UAS systems where a human operator authorizes each engagement fit the HITL model directly. The event-camera advantages — low alert latency, reduced operator workload, better degraded-condition performance — are all directly relevant.

The review’s identification of the HITL dataset gap also points to a meaningful research need: collecting data that includes human decision times and accuracy in C-UAS HITL scenarios with event-camera sensors.

---

Applications of Neuromorphic/Event Camera in Robotics With Human in Loop: A Systematic Review, Datasets, and Challenges. IEEE Xplore, 2025.